Microhabitat Analysis Using Radiotelemetry Locations and Polytomous Logistic Regression

Microhabitat analyses often use discriminant function analysis (DFA) to compare vegetation structures or environmental conditions between sites classified by a study animal's presence or absence. These presence/absence studies make questionable assumptions about the habitat value of the comparison sites and the microhabitat data often violate the DFA's assumptions of an equal covariance structure and multivariate normality. An alternative is to generate an ordinal measure of site-use intensity from radiotelemetry locations. This measure is derived from the percentage of total telemetry points of a study animal that are found at use-only sites, overcoming many of the problems associated with defining "absence" sites. The use-intensity response is then modeled as a function of microhabitat variables using ordered polytomous logistic regression (PLR). Unlike DFA, PLR does not require covariance equality or multivariate normality, and allows categorical microhabitat variables. The classification error of the microhabitat model developed with PLR is then assessed by jackknifing. This technique is demonstrated with an example analysis of the foraging microhabitat of the northern spotted owl (Strix occidentalis caurina). The resulting model correctly classified 78% of the sample stands in the jackknife evaluation. For animals with site fidelity and radiotelemetry data, the proposed technique may provide a robust alternative for microhabitat analysis.